Viet-Anh on Software Logo

What is: Corner Pooling?

SourceCornerNet: Detecting Objects as Paired Keypoints
Year2000
Data SourceCC BY-SA - https://paperswithcode.com

Corner Pooling is a pooling technique for object detection that seeks to better localize corners by encoding explicit prior knowledge. Suppose we want to determine if a pixel at location (i,j)\left(i, j\right) is a top-left corner. Let f_tf\_{t} and f_lf\_{l} be the feature maps that are the inputs to the top-left corner pooling layer, and let f_t_ijf\_{t\_{ij}} and f_l_ijf\_{l\_{ij}} be the vectors at location (i,j)\left(i, j\right) in f_tf\_{t} and f_lf\_{l} respectively. With H×WH \times W feature maps, the corner pooling layer first max-pools all feature vectors between (i,j)\left(i, j\right) and (i,H)\left(i, H\right) in f_tf\_{t} to a feature vector t_ijt\_{ij} , and max-pools all feature vectors between (i,j)\left(i, j\right) and (W,j)\left(W, j\right) in f_lf\_{l} to a feature vector l_ijl\_{ij}. Finally, it adds t_ijt\_{ij} and l_ijl\_{ij} together.